Median¶
A central cut of ordered data or a distribution with at least half the observations or probability at or below it and at least half at or above it.
Core Idea¶
A median is a central cut of ordered observations or a probability distribution: at least half lie at or below it and at least half at or above it. For odd-sized samples the middle ordered value is the usual result. For even-sized numerical samples, averaging the two central values is a common convention, though every point between them satisfies the half-mass condition.[^ref-54dddd583e72]
Scope of Application¶
The median summarizes central rank in a skewed or heavy-tailed sample. NIST's concrete 10,000-value Cauchy illustration reports mean 3.70, median −0.016 and extremes near −29,000 and 89,000; those are one generated sample's values, not population constants or income data. SciPy documents filtering its ascent image with a size-20 neighborhood and chooses sorted index \(n//2\), rather than an even-window midpoint. These cases show both the rank mechanism and the importance of the reporting convention.[ref-54dddd583e72][ref-a5e456a452a9]
Clarity¶
For \(1,2,3,4,100\), the median is \(3\); replacing \(100\) with \(1{,}000{,}000\) does not move it. For \(1,2,4,9\), the midpoint convention reports \(3\), but any value between \(2\) and \(4\) meets the median condition. Ties may place more than half the observations at the median rather than exactly half strictly on each side.[^ref-54dddd583e72]
Manages Complexity¶
One central value resists a small number of arbitrarily extreme observations and is the half-level member of the quantile family. It omits tail shape and may still be distorted if enough observations change; the familiar breakdown point is near one-half, not an absolute immunity claim.[^ref-04fc184614e2]
Abstract Reasoning¶
Specify the ordered carrier and whether observations have weights. Sort finite data or examine the distribution's cumulative mass; test both weak half-mass inequalities; then state how a single value was chosen if the median is nonunique. Absolute-deviation minimization is a consequence for finite numerical samples, not the definition for all ordered categories.
Knowledge Transfer¶
The same central-order rule applies to incomes and pixel windows, while population weights, image boundaries and implementation conventions differ. The skeleton is an ordered half-mass cut; this statistical functional is not a domain-neutral prime. Order is a strict prerequisite because the two weak half-mass inequalities require a ranking relation. The current Quantile entry uses a generalized-inverse endpoint, so it is related but does not strictly subsume every valid point of a nonunique median interval. “Median-unbiased estimator” is a different estimator property, not an alias of Median.
[^ref-54dddd583e72]: NIST/SEMATECH Engineering Statistics Handbook, Measures of Location. Odd/even sample convention and comparison with the mean. [^ref-a5e456a452a9]: SciPy maintainers, median filter documentation. Neighborhood median and exact even-size implementation behavior. [^ref-04fc184614e2]: Rousseeuw's first-hand explanation of the sample median's breakdown intuition. “Slightly less than half” qualification.
Relationships to Other Abstractions¶
Current abstraction Median Domain-specific
Parents (1) — more general patterns this builds on
-
Median presupposes Order Prime
Median's weak half-mass inequalities and central ranks require an order relation on the carrier.
Hierarchy paths (3) — routes to 3 parentless roots
- Median → Order → Comparison → Self Checking
- Median → Order → Relation
- Median → Order → Set and Membership
Neighborhood in Abstraction Space¶
Median sits in a sparse region of the domain-specific corpus (86th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Descriptive Statistics & Correlation Measures (39 abstractions)
Nearest neighbors
- Unimodality — 0.82
- Stochastic ordering — 0.82
- Five-number summary — 0.81
- Bounded complete poset — 0.81
- Quantile — 0.80
Computed from structural-signature embeddings · 2026-10-08